13 Dynamic Contrast-Enhanced MR Imaging in Musculoskeletal Tumors

(1)

13 Dynamic Contrast-Enhanced MR Imaging in Musculoskeletal Tumors

June S. Taylor and Wilburn E. Reddick

J. S. Taylor, PhD

Associate Professor, Department of Radiology, University of Utah, Utah Center for Advanced Imaging Research, 729 Arapeen Drive, Salt Lake City, UT 84108, USA

W. E. Reddick, MD, PhD

Assistant Member, Department of Diagnostic Imaging, St. Jude Children‘s Research Hospital, 332 North Lauderdale Street, Memphis, TN 38105-2794, USA

Dedication

In memory of Charles B. Pratt, MD (1930–2002), a rare combination of gentleman, scholar, and respected practitioner of oncology. He cared for his patients and his profession, and took the time to teach us why.

13.1 Introduction

13.1.1 Four Oncologic Questions

Radiologists are required to provide information on four clinical questions in imaging primary muscu- loskeletal neoplasms: diagnosis, staging, response to pre-surgical (or initial) chemotherapy

¹

, and detec- tion of recurrence. MRI is useful, together with radi- ography and clinical data, in narrowing the differen- tial diagnosis and establishing the extent of disease.

However, it rarely provides a deﬁ nitive diagnosis for these tumors. MRI is the pre-eminent imaging modality for the remaining three clinical questions (Bloem et al. 1997).

MRI has dramatically improved the pre-operative staging of these tumors. As a result, there has been a major reduction in the incidence of mutilating sur- gery. Limb-sparing procedures are now commonly used for primary osseous or soft-tissue neoplasms, with resulting improvement in the quality of life for these survivors. Patients with primary osseous and soft-tissue neoplasms have also beneﬁ ted enormously from developments in initial chemotherapies. Initial chemotherapy is administered prior to tumor resec- tion and has led to improved surgical outcomes and improved survival. For some tumors and/or stages, response to this initial therapy correlates positively with overall survival, in which case measuring this response as precisely and accurately as possible is vital.

MRI is also important in screening for the pres- ence of residual or recurrent tumor. The detection of small recurrent or residual tumor, following initial treatment, poses a challenge. The problem is fur- ther complicated by susceptibility artifacts on CT or MR images arising from reconstructive hardware or metallic particles from surgical instruments. In spite

CONTENTS

Dedication 215 13.1 Introduction 215

13.1.1 Four Oncologic Questions 215

13.1.2 MR Imaging of Bone and Soft-Tissue Tumors 216 13.1.3 Rationale for DCE-MRI in Advancing Treatment

of Musculoskeletal Tumors 216 13.1.3.1 Present Applications 217 13.1.3.2 Future Directions 218 13.2 Primary Osseous Sarcomas 218 13.2.1 Pediatric Osteosarcoma 218

13.2.1.1 Rationale for DCE-MRI of Osteosarcoma 219 13.2.1.2 Acquisition and Analyses

of DCE-MRI of Osteosarcoma 220 13.2.1.3 Clinical Application of DCE-MRI

to Assess Osteosarcoma Response 223 13.2.1.4 Clinical Application of DCE-MRI

to Predict Disease-Free Survival 224 13.2.2 Ewing’s Family Tumors (EFTs) 225 13.2.2.1 Rationale for DCE-MRI of EFT 226 13.2.2.2 Clinical Application of DCE-MRI

to Assess EFT Response 227 13.2.3 Cartilaginous Tumors:

Enchondroma vs. Chondrosarcoma 228 13.3 Primary Soft Tissue Sarcomas 229

13.3.1 Clinical Application of DCE-MRI to Distinguish Malignant from Benign Soft Tissue Tumors 229 13.3.2 Potential Clinical Application of DCE-MRI

in Soft Tissue Sarcoma 230 13.3.2.1 Response 230

13.3.2.2 Follow-Up 231 13.4 Synopsis 231 References 234

1 Pre-surgical chemotherapy is sometimes referred to as induction or neoadjuvant therapy.

JACK_13-Taylor.indd 215

JACK_13-Taylor.indd 215 19.10.2004 12:13:57 Uhr19.10.2004 12:13:57 Uhr

(2)

of this, MRI has proved to have superior sensitivity and speciﬁ city to CT in the detection of recurrent musculoskeletal tumor (Berger et al. 2000; Bloem et al. 1997). In patients who have received radiation therapy, radiation-induced soft tissue edema may arise and persist for years. In such cases, dynamic contrast-enhanced MRI (DCE-MRI) may be useful in discriminating between radiation effects, which enhance more slowly, and recurrent tumor, which has a much faster rate of contrast uptake.

13.1.2 MR Imaging of Bone and Soft-Tissue Tumors

A broad consensus on diagnostic assessment of mus- culoskeletal tumors has emerged over the past decade.

Radiographs, clinical data and MRI can narrow the differential diagnosis, and in some cases precisely diag- nose the tumor type (Ma 1999; van der Woude et al. 1998a). However, biopsy is still required for accu- rate diagnosis of musculoskeletal lesions. Protocols for MR imaging and the intricacies of interpretation in diagnosing musculoskeletal lesions have been well described in recent reviews (van der Woude et al.

1998a; Ma 1999; Berger et al. 2000). Table 13.1 shows a basic protocol for imaging musculoskeletal tumors that follows current recommendations (Ma 1999; Berger et al. 2000). Fat suppression is necessary to improve discrimination between fatty marrow components and tumor or red marrow; it improves the sensitivity of detecting edema and ﬂ uid collections in the marrow and soft tissues. Some prefer short-tau inversion recover (STIR) images to fat-suppressed T2-weighted

spin-echo images, because STIR fat suppression is less susceptible to the effects of local ﬁ eld inhomogeneities;

on the other hand, the SNR of STIR images is lower.

This choice is also inﬂ uenced by the ﬁ eld strength and other characteristics of the MR imaging equipment.

There has been disagreement among musculo- skeletal radiologists on the necessity for contrast administration in MRI examinations of musculoskel- etal tumors (Berger et al. 2000; van der Woude and Egmont-Petersen 2001) . A case can be made that neither static nor dynamic contrast-enhanced imag- ing is critical to the diagnosis of these lesions (van der Woude and Egmont-Petersen 2001), because the use of contrast has not been shown to improve the discrimination between benign and malignant tumors, except in cartilaginous tumors. There- fore biopsy is required for any lesions suspicious for malignancy. However, contrast administration should be planned for examinations done for staging, for measuring response to therapy, and for follow-up.

DCE-MRI is more discriminating than static contrast MRI for the staging of musculoskeletal lesions, and DCE-MRI may also be necessary to guide the biopsy for optimal tumor sampling. Unless malignancy can be ruled out deﬁ nitively (as for some bone and soft tissue cysts, lipomas and hemangiomas), it is prudent to plan the use of DCE-MRI in the staging examina- tion of these tumors.

For staging of primary neoplasms of bone and soft tissue, and for surgical planning, including biopsy, MRI is the imaging examination of choice (van der Woude and Egmont-Petersen 2001). It is superior in determining the extent of tumor and the tumor’s relation to adjacent muscles and neurovascular structures. It is thus invaluable in planning biopsies and surgical treatment, both of which are critically important in many of these lesions. Staging and treat- ment of musculoskeletal lesions have progressed to the point that limb-sparing surgery for extremity lesions is frequently undertaken. Biopsies of possibly malignant musculoskeletal tumors are planned with two priorities in mind: ﬁ rst, to obtain the most accu- rate diagnosis, and second, to avoid compromising the future treatment and prognosis of the patient.

13.1.3 Rationale for DCE-MRI in Advancing Treatment of Musculoskeletal Tumors

DCE-MRI generally refers to a series of T1-weighted images acquired before, during and after contrast injection. It has been successfully performed on

Table 13.1. Representative MR imaging procedure for muscu- loskeletal tumors

Contrast Orientation Sequence Contrast

medium weighting

Pre Axial Spin echo T1

Pre Axial Fast spin echo T2 with Fat Sat^a Pre Longitudinal^b Spin echo T1

Pre Longitudinal Fast spin echo T2 with Fat Sat

and/or STIR^c

Post Axial Spin echo T1 with Fat Sat Post Longitudinal Fast spin echo T1

a Fat Sat: fat saturation with chemical-shift-selective presatu- ration RF pulse; this method cannot be used when prostheses are present.

b Longitudinal: coronal or sagittal plane along long axis of bone, including nearest joint in FOV.

c STIR: short-tau inversion recovery; also suppresses fat signal.

(3)

imaging systems with ﬁ elds from 0.5–1.5 T, although higher ﬁ eld strengths clearly are preferable to achieve the best temporal and spatial resolution of tumor.

Historically, the data have been collected by 2D spin- echo or gradient-echo imaging of single or multiple slices (Verstraete et al. 1994b; Hanna et al. 1992;

Reddick et al. 1996), and more recently by 3D spoiled gradient echo imaging (Knopp et al. 2001). 3D imag- ing has the advantage of more complete coverage of tumor, although temporal sampling times short enough for the question addressed must be main- tained. It may also be difﬁ cult to include a feeding artery in the 3D slab for simultaneous determina- tion of the dynamic (intensity vs. time) curve in the blood, which improves reproducibility (Rijpkema et al. 2001). The temporal sampling required is dis- cussed below for speciﬁ c applications.

13.1.3.1

Present Applications

Contrast-enhanced MRI, particularly DCE-MRI, is the method of choice for monitoring response to pre-operative chemotherapy in certain malignant musculoskeletal tumors. DCE-MRI is broadly useful for detecting residual or recurrent tumor after sur- gery (van der Woude and Egmont-Petersen 2001;

Ma et al. 1997; Verstraete and Lang 2000). The experience with DCE-MRI for staging local extent of these tumors and for planning biopsy procedures are discussed below under speciﬁ c families of mus- culoskeletal tumors.

²

The earliest successful application of DCE-MRI to oncologic imaging was in assessing osteosarcoma response to therapy (Erlemann et al. 1989; Fletcher 1991; Bonnerot et al. 1992). Primary bone tumors, in particular, provide a near-ideal system for evidence- based imaging response studies, because at the end of initial chemotherapy the tumor is resected en bloc and the ground truth for tumor response is available from histological analyses of slides matched to the imaging plane(s) (Egmont-Petersen et al. 2000;

Hanna et al. 1993). In the decade and a half since the initial results, numerous studies have conﬁ rmed the efﬁ cacy of DCE-MRI in assessing tumor necro- sis in osteosarcoma. There is enormous variability in tumors and therefore it is important to know how best to utilize DCE-MRI for particular clinical ques-

tions. Researchers have obtained promising results on the usefulness of DCE-MRI in assessing response to initial chemotherapy for other, rarer musculoskel- etal tumors, speciﬁ cally Ewing’s family tumors and soft-tissue sarcomas.

Malignant musculoskeletal tumors resemble normal bone or soft tissues in immunogenic and other properties, but their architecture is different in key ways that can be exploited by DCE-MRI. Their growth and development are chaotropic in varying degrees. This leads to abnormalities in the archi- tecture and permeability of their microvasculature (Jain 1994, 1996 b), and sometimes to remarkable spatial heterogeneity in perfusion of tumor cells.

Also, the interstitial space (the extracellular-extra- vascular space, or EES, compartment) of malignant tumors may be many times larger than normal or even edematous tissues (Gullino and Grantham 1964) , so that uptake into the EES compartment is often a signiﬁ cant part of enhancement in tumors.

The distribution of this EES is often inhomogeneous, with some tumor regions having expanded EES and others having a more normal EES.

DCE-MRI can be used to assess the microcircu- lation and EES of malignant bone sarcoma. It pro- vides a way to discriminate between the uptake of contrast into necrotic or edematous tissue and the leakage of contrast agent from angiogenic microvas- culature into the EES of viable tumor. It does this by using either semi-quantitative or quantitative mea- sures of contrast uptake. Separating viable tumor from necrotic tumor and edema is valuable in the staging and biopsy of these lesions. Identifying the fraction of dead tumor just before tumor resection gives a measure of response to initial chemotherapy, and response to therapy has predictive value for over- all survival, particularly in patients with local (M0) disease. Moreover, the rate of contrast uptake into various tissue regions appears to serve as a surro- gate for the access of drug to these regions, at least in certain tumor types. Basic research in animal tumor models has shown that transient irradiation-induced increases in tumor capillary permeability to cisplatin can be quantiﬁ ed with DCE-MRI (Schwickert et al. 1996). More recent animal studies of solid tumor response to chemotherapy (Su et al. 1999) and to gene therapy (Su et al. 2000) have demonstrated the abil- ity of DCE-MRI to detect vascular changes correlated with response or progression. In clinical observations of numerous series of patients with bone and soft tissue sarcoma, measures of contrast uptake (contrast access) have convincingly demonstrated a relation- ship with measures and predictions of the tumor’s

2 The staging of musculoskeletal lesions is complex and, with respect to soft-tissue sarcomas, a work in progress. The inter- ested reader is referred to articles by (PEABODY et al. 1998;

Gebhardt 2002).

(4)

response to preoperative chemotherapy. The results of these studies have indicated that greater access at the time of presentation, greater decrease in access during therapy, and low access at the completion of preoperative therapy correspond to better response and longer disease-free survival (Verstraete and Lang 2000; Reddick et al. 2001). Consequently, DCE- MRI studies of pre-operative response to therapy should be considered for every patient with a che- motherapy-sensitive musculoskeletal malignancy, for the purposes of adjusting the plan of pre-surgical chemotherapy, revising surgical planning and choos- ing optimal post-surgical therapy.

13.1.3.2

Future Directions

For chemotherapy to be effective, therapeutic agents must reach target cells (tumor or endothelial) in adequate concentrations and with minimal toxicity to normal tissues (Jain 1996a). Once a blood-borne molecule reaches an exchange capillary, transport across the vessel wall is a function of the surface area of exchange, the transvascular concentration gradi- ent and the interstitial ﬂ uid pressure. Tumor intersti- tial pressure can be much higher than that in normal tissues, to the point of collapsing perfusing microves- sels. It was hypothesized that, because of aberrations in the architecture of microvessels and interstitial pressure, some solid tumors may resist drug penetra- tion (Jain 1994). Such physiological resistance may play a signiﬁ cant role in treatment failure (Tannock et al. 2002; Jain 2001), and is a different mechanism of resistance to therapy from the better-understood cellular resistance phenomenon conferred by the presence of resistance molecules, such as the mul- tiple drug resistance protein Mdr1, in tumor cells.

However, it has been difﬁ cult to test this hypothesis of physiological resistance in clinical studies, in the absence of non-invasive repeatable measures of drug penetration that sample tumor compartments other than plasma. Now that advances in tumor biology and intelligent drug design are producing many novel therapeutic agents for clinical trials, it is increasingly important to discriminate between cellular resistance of tumor cells to drug and physiological resistance arising from a failure to expose poorly-perfused tumor regions to sufﬁ cient concentrations of drug.

This concern is driving the interest of researchers in DCE-MRI as a methodology which can supply new, urgently-needed pharmacokinetic information on the efﬁ ciency and heterogeneity of drug access to solid tumors in individual patients.

Optimizing the spatial and temporal sampling rates for DCE-MRI applications is complex, because faster image acquisition rates increase the temporal resolu- tion of the dynamic curve (time-intensity curve), but the signal-to-noise ratio (SNR) and/or spatial reso- lution of each image is decreased. Likewise, there is the usual trade-off between SNR and better spatial resolution (voxel size) in the DCE images. Current protocols for 2D and 3D DCE-MRI have been pub- lished by several European researchers: Shapeero and Vanel (2000); Rijpkema et al. (2001) for 2D and Knopp et al. (2001) for 3D. However, the DCE-MRI protocol should begin prior to contrast injection, as emphasized by Knopp et al. (2001). The sequence used should give tissue signal intensity proportional to contrast concentration in the tissue.

It is important to recognize that reduction of the SNR below a certain threshold will be detrimental to the reliability and sensitivity of the calculations of DCE-MRI measures. Noisy dynamic curves lead to decreases in reliability of both empirical and model estimates, unless methods that are robust to noise are employed. Analyses by region of inter- est (ROI), which averages together the information from many pixels, can be used to obtain dynamic (time vs. intensity) curves that have higher SNR than curves from individual pixels. This trading of spatial information for SNR is not always possible for all tumors in all applications. A rich area for research is the development of algorithms that can cope with noisy data while maintaining the spatial resolution of DCE images.

13.2 Primary Osseous Sarcomas

13.2.1 Pediatric Osteosarcoma

Osteosarcoma is the most common malignant bone tumor of childhood (60%) (Link and Eilber 1993).

Current treatments for these tumors consist of sev- eral cycles of initial chemotherapy given before sur- gical ablation of the osteosarcoma for local control (Fletcher 1997). Preoperative chemotherapy is used to treat systemic disease (micrometastases), provide time for planning surgical therapy (including manu- facturing customized prostheses), and allow time for postoperative healing before postoperative chemo- therapy is given (Link et al. 1986; Rosen et al. 1979).

It is important to note that the period of preoperative

(5)

chemotherapy can be used to evaluate new chemo- therapeutic agents in phase 2 trials.

The use of preoperative chemotherapy has pro- duced a dramatic improvement in prognosis for chil- dren with bone sarcoma (Link et al. 1986). Recent studies show at least 70% of children with osteosar- coma now survive (Grier 1997). Accurate determina- tion of response is essential in assessing the effective- ness of investigational and standard drug therapies during the pre-surgical treatment window.

There is a paucity of data on primary osteosar- coma in adults. While clinical, histopathologic and prognostic features are well understood for children (under 20) with osteosarcoma, much less is known about these factors for adults. There is a tendency for axial location of primary osteosarcoma in adults (Huvos 1986; Carsi and Rock 2002), compared to children. The distribution of histologic variants in adults is similar to those seen in pediatric osteosar- coma (Carsi and Rock 2002). Negative prognostic factors appear to be metastases prior to or during treatment, large tumor volumes, a pathologic fracture prior to treatment, and inadequate surgical margins.

These are not different in kind from negative factors that are known for pediatric osteosarcoma. One pos- sibility is that the poorer outcome in treating pri- mary osteosarcoma in adults is related to the higher fraction of adults presenting with large, advanced stage tumors and the greater fraction of axial tumors, which are much harder to resect cleanly. Reports to date on the response rate for adults to chemotherapy are conﬂ icting (Antman et al. 1998; Bacci et al. 1998;

Bielack et al. 2002). Results of older treatment regi- mens in adults with primary osteosarcoma have been poor (15%−40% survival rate), and prospective trials of more aggressive multi-agent and multi-modality protocols are needed. European trials have provided evidence of better survival for adults treated with therapy protocols similar to those used in children (Bacci et al. 1998; Bielack et al. 2002).

13.2.1.1

Rationale for DCE-MRI of Osteosarcoma

Osteosarcoma, like all malignant musculoskeletal tumors, requires biopsy for deﬁ nitive diagnosis.

DCE-MRI is useful for staging and for optimal tar- geting of tumor biopsy in these large, heterogeneous tumors (Geirnaerdt et al. 1998; van der Woude et al. 1998b).

Evaluating response of bone sarcoma to initial che- motherapy by imaging methods is a challenge. Even osteosarcoma that have become largely necrotic may

not shrink signiﬁ cantly, perhaps because of their usu- ally extensive osteoid and bony matrix (Wellings et al. 1994). At present, the gold standard for assessing the effects of chemotherapy is histological examina- tion of a representative section of the resected tumor (Huvos et al. 1977; Picci et al. 1985; Salzer-Kunts- chik et al. 1983; Raymond et al. 1987). Response is considered good if there is at least 90% tumor cell necrosis (Rosen et al. 1982). However, some centrally located tumors may not be resectable and therefore cannot be evaluated by histology.

Tumor necrosis resulting from initial chemotherapy is strongly associated with effective local control. The degree of response is an important prognostic factor that can be used to plan post-surgical treatment and to optimize the timing of surgery (Raymond et al. 1987;

Glasser et al. 1992; Rosen et al. 1979, 1982; Picci et al.

1994; Hudson et al. 1990). Response is also an impor- tant determinant of the patient’s eligibility for limb-sal- vage procedures without increased risk of local recur- rence (Picci et al. 1994; Gherlinzoni et al. 1992).

This section will focus on two vital issues in the treatment of children with primary osteosarcoma:

(1) how to detect response of the primary tumor to initial chemotherapy, and (2) how to identify at initial staging those tumors that are likely to fail pre-surgical therapy. A ﬁ rst approach to answering the response question requires comparing imaging methods with histology. To validate the imaging method for grading response requires the additional step of comparing both histological and imaging methods to the most important outcome measure, disease-free survival, in the same population. Likewise, success in predict- ing the risk of treatment failure at diagnosis requires, ﬁ rst, comparison of the predictive imaging measure with histopathology and imaging at the end of the initial chemotherapy window, and for full validation, comparison with patient survival.

No satisfactory clinical or static imaging meth- ods exist for determining response of osteosarcoma to cytotoxic drugs before resection (Erlemann et al. 1990). Originally, conventional static MR imaging showed promise in response assessment (Holscher et al. 1990, 1992; Pan et al. 1990). However, MR signal intensities of viable and necrotic tumor, edema, and hemorrhage overlap on T2-weighted images (Pan et al. 1990; Sanchez et al. 1990). In addition, the use of STIR and contrast-enhanced T1-weighted MR imaging overestimates the extent of viable tumor because of enhancement of nonmalignant reactive tissue (Erlemann et al. 1990; Onikul et al. 1996;

Glazer et al. 1985). A large study of the effective- ness of CT scanning and conventional MR imag-

(6)

ing for evaluating primary tumor response on the basis of several criteria showed that the ability of these imaging modalities to predict histopathology response and outcome was limited (Lawrence et al.

1993) .

Other imaging methods that have been used to assess response in bone sarcoma before sur- gery include conventional radiography, computed tomography, radionuclide scintigraphy, angiogra- phy, and ultrasonography, as well as conventional static MR imaging. All proved less than satisfac- tory in measuring response, as summarized in a recent review (Fletcher 1997). On the other hand, there are now multiple clinical studies from several groups on DCE-MRI for measuring response to ini- tial chemotherapy, and these collectively show that DCE-MRI measures of contrast agent uptake into osteosarcoma provide a reliable, accurate measure of response.

This agreement is satisfying. However, a variety of semi-quantitative and quantitative measures were extracted from the DCE-MR images to arrive at these results, and the multiplicity of measures has created confusion. It is worth comparing the DCE-MRI mea- sures applied to assess response in osteosarcoma studies, as a basis for discussing the pitfalls in gen- eralizing these to other musculoskeletal tumors and other oncologic questions.

13.2.1.2

Acquisition and Analyses of DCE-MRI of Osteosarcoma

Osteosarcoma occur most frequently in the long bones of the extremities and expand in both the intra- medullary cavity and the surrounding soft tissue.

Therefore, the localization of the tumor is somewhat constrained by the bone. DCE-MRI acquisitions in a single 10-mm thick plane with in-plane resolution of approximately 3 mm

²

can assess response and iden- tify foci of residual tumor with parametric maps, which can be directly compared with histologic maps.

Both measuring response and predicting response to therapy prospectively may be done by assessing a single coronal slice, as is the practice in histologic measures, but prediction in particular may prove most accurate when the entire tumor is sampled. For best discrimination, the choice of a temporal sam- pling rate should adequately sample the fastest accu- mulation rate. In some of the best-perfused regions of an osteosarcoma, uptake of the contrast agent can equilibrate in as little as 25–30 s; thus, a minimum sampling rate of one image every 13–15 s is required

for osteosarcoma. In order to evaluate patient image data using a kinetic model with a plasma term (arte- rial input function, or AIF) the DCE-MRI would have to sample the dynamic curve at approximately one image every 3 s.

Until recently, the majority of clinical investiga- tions have analyzed DCE-MRI by extracting a variety of empirical parameters as measures. The majority of radiologists must still do so, lacking the expert software to extract the contrast medium concentra- tion–time curve from the DCE-MRI images. The semi-quantitative analyses calculate or estimate parameters that are related to this concentration–

time curve: the absolute change in enhancement over some time period, or the relative change (divided by the baseline intensity before contrast injection), or the rate of relative (or absolute) enhancement in the initial period (ﬁ rst 30−60 s) after the contrast injec- tion, or in some few cases the delay in contrast reach- ing the ROI. The enhancement may or may not be normalized to baseline intensity or compared with other normal tissues in the imaging plane. One of the most popular intensity-based methods takes the ratio of the relative signal intensity enhancement over the time required for the enhancement, to calculate a rate of enhancement (Slope and percentage Slope). This accumulation rate, if it is computed over a short time period (<30 s for osteosarcoma), provides a measure of the rate of contrast agent accumulation in the vas- culature and tumor interstitial spaces. Another mea- sure is the maximum enhancement that occurred at any point in the signal intensity curve. The maximum enhancement provides a measure of the total accumu- lation of contrast agent in the tumor vasculature and interstitial spaces, and this is related to the maximum concentration of contrast agent accumulated in these tumor spaces. Some studies showed it was useful to combine both these measures into a lumped measure, the dynamic magnitude vector (DVM), which is a weighted sum reﬂ ecting both the maximum magni- tude and the initial rate of enhancement.

More recently, researchers began extracting the contrast medium concentration–time curve from DCE-MR images. This allowed them to apply estab- lished compartmental (pharmacokinetic) methods to model k

ep

, the ﬂ ux rate constant between EES and plasma, for low-molecular weight (MW) contrast from plasma to tumor EES (related to the slope mea- sures) and the extent to which these agents accu- mulate in the EES space of the tumor (related to the maximum enhancement measures). Although very complex models can be developed to describe indica- tor distribution, a simpler two-compartment model

(7)

may be more appropriate to the limited SNR and spatial resolution of the DCE-MRI images. A con- ventional two-compartment pharmacokinetic model consisting of plasma volume and EES volume has been used in most studies that have analyzed the dis- tribution of contrast in each image voxel. Tofts et al.

(1999) have reported a set of standardized quantities and symbols to be used in estimating kinetic param- eters from DCE-MRI (Tofts et al. 1999). The most common parameter reported in the literature is the k

ep

. As we will show below, the initial rate measures can provide a more or less accurate approximation to the pharmacokinetic parameter k

ep.

To demonstrate the enhancement–time curves seen, an actual clinical osteosarcoma case (Fig. 13.1) shows these curves for regions seen in a very large tumor of the distal right femur, which was resected and sectioned in the same plane as the MRI studies, for histopathology correlation. According to histo- pathology, this tumor responded very poorly to pre- surgical therapy, with less than 10% necrosis at resec- tion. The characteristic shape of the dynamic signal after the contrast arrives at the tumor is shown for four different regions of the tumor (taken from DCE- MR images just prior to surgery). The ﬁ rst region of the tumor (A) is from the central core, which in this tumor is avascular and necrotic. The lack of signal change during DCE-MRI precludes assessment of the pharmacokinetic model variables for this region.

However, this characteristic signal pattern identi- ﬁ es the region as necrotic. All of the three remain- ing regions contained some viable tumor by histol- ogy. In the second region (B), an area of soft tissue

adjacent to the central necrotic core, the MR signal is low in intensity and slow to enhance. This region corresponds to a semi-necrotic, edematous section of the tumor, which also includes reactive tissues. Phar- macokinetic modeling of the MR signal in this region demonstrates a relatively small transfer rate constant, k

ep,

which would indicate moderate access to the EES compartment. However, the contrast continues to accumulate throughout the 6 min of the DCE-MRI acquisition, most likely due to an expanded EES compartment. The third tumor region (C) shows more rapid and pronounced enhancement than does region B, as reﬂ ected by a larger value of k

ep

. This well- perfused region of viable tumor with stable microcir- culation demonstrates good access and accumulation of contrast agent. Unlike region B, the enhancement (and therefore the contrast concentration) in region C stabilizes within about 2 min of contrast arrival at the tumor and remains fairly constant during the remainder of the acquisition. The fourth region (D) has a very rapid increase in contrast concentration during the 30 s following contrast arrival, and a large degree of enhancement at the end of the measure- ment. Pharmacokinetic analysis demonstrates a very large k

ep

for D. The size of this rate constant indicates rapid exchange between the contrast in the vascular and tumor EES compartments, so that the resulting concentration of contrast agent in tumor EES in this region closely follows the plasma concentration. This behavior is characteristic of rapidly proliferating tumor.

Analyzing regions B through D in Fig. 13.1 with each of the techniques listed in Table 13.2 provides

Fig. 13.1. T1-weighted MR imaging of distal femur osteosarcoma after injec- tion of contrast, status post-initial chemotherapy and prior to tumor resection.

Regions of representative tissue types are shown with corresponding dynamic signals for four regions of the tumor: A, necrotic central core; B, viable soft tissue component with increased EES; C, viable marrow with stable microcir- culation; D, rapidly proliferating region

(8)

measures extracted from the same DCE-MRI exami- nation of the osteosarcoma of Figure 13.1. These parametric images demonstrate that each analysis technique provides somewhat different information, depending on the individual measure’s sensitivity to characteristics of the dynamic concentration–time curve. Generally, all three measures increase in value from region B to region C to the rapidly proliferat- ing region D. However, the combined measure DVM is slightly higher in B than C primarily due to the continued accumulation of contrast in B. All three measures show that the viable tumor regions have at least moderate access to contrast agent, indicat- ing a good blood supply and effective transfer from plasma to EES for agents of low molecular weight (like GdDTPA)

Table 13.2. Results of three different data analysis techniques [% slope, dynamic vector magnitude (DVM), and pharmacokinetic modeling) for the dynamic signals from regions B (soft tissue region – semi-necrotic with expanded EES), C (marrow region – stable microcirculation), and D (rapidly proliferating region – leaky neovasculature) in Fig. 13.1

Region B Region C Region D

Soft Tissue Marrow Rapid

Proliferation

% Slope (%/min) 32.55 54.92 81.87

DVM (SI/s) 4.24 4.09 7.17

k_ep (min^-1) 1.06 1.55 2.58

a practical comparison of one pharmacokinetic and two empirical measures. Figure 13.2 demonstrates the parametric images generated from each of three

Fig. 13.2a-d. T1-weighted MR imaging of distal femur osteosarcoma after injection of contrast (a). Entire tumor is selected as the region of interest. Parametric images calculated from the analysis of the DCE-MRI signals on a pixel-by-pixel basis are shown:

b, percent slope image; c, dynamic vector magnitude (DVM) image; d, pharmacokinetic modeling image of k_ep

a b c d

(9)

13.2.1.3 Clinical Application of DCE-MRI to Assess Osteosarcoma Response

Traditionally, studies performed to determine the response of bone sarcoma to preoperative chemo- therapy have concentrated primarily on the early contrast uptake phase (initial slopes, k

ep),

because this can be understood as an assessment of microcircula- tion. The results of many of these studies have been summarized in Table 13.3 for easy comparison. The clinical application of DCE-MRI in bone sarcoma began with the fi nding by Erlemann et al. (1989) that necrotic areas within tumors enhance less rap- idly and less intensely than viable tumor. The DCE- MRI images were analyzed on the basis of the differ- ences in percent slope before and after chemotherapy from a circular ROI encompassing the region exhib- iting maximal signal increase on the presentation examination. Within a year, Fletcher et al. (1992) published an independent verifi cation of these earlier results, using one or more ROIs of various shapes and sizes to allow sampling of as much of the enhanc- ing portion of the tumor as possible. Concurrently, Hanna et al. (1992) reported a refi nement of the ear- lier techniques that increased the spatial resolution by using small contiguous regions, followed later by pixel-by-pixel mapping (Hanna et al. 1993) to detect small foci of residual tumor and to compensate for regional variations in tumor perfusion.

The percent slope analysis approach continues to be used for assessing tumor response. Recently, both Kawai et al. (1997) and Ongolo-Zogo et al. (1999) reported results of DCE-MRI of bone tumors ana- lyzed on the basis of the differences in percent slope before and after chemotherapy, and the results were compared with the histological responses.

Building on evidence that the early initial slope of the DCE-MRI signals provided some surrogate assessment of response in bone sarcoma, Reddick et al. developed a more complete characterization of the DCE-MRI enhancement-time curves (Reddick et al. 1994) and applied this method to studying response in patients with osteosarcoma (Reddick et al. 1996). Three useful variables were identiﬁ ed:

the initial contrast accumulation rate (ICAR), the delayed contrast accumulation rate (DCAR), and the maximum enhancement (ME) reached in the dura- tion (6.5 min) of the DCE-MRI study. Most of the information appeared to lie in the initial rate and maximum enhancement parameters. This lead to creating a single measure, the dynamic vector mag- nitude (DVM), which combined the information on how much (ME) contrast entered the region and how rapidly (ICAR) it did so.

An analytic method borrowed from nuclear medi- cine, factor analysis of medical image sequences (FAMIS), was used by Bonnerot et al. (1992) in a study of ten osteosarcoma patients. FAMIS pro-

Table 13.3. Predicted DCE-MRI accuracy compared to histological grading of response of bone sarcoma to preoperative che- motherapy

DCE-MRI measure Critical value Accuracy^a Sensitivity Speciﬁ city

(non-response) (response)

Erlemann et al. (1990) % Slope from 60% 86% 80% 91%

linear estimate reduction (18/21) (8/10) (10/11)

Bonnerot et al. (1992) Factor analysis No vascular 90% 80% 100%

(FAMIS) phase (9/10) (4/5) (5/5)

Fletcher et al. (1992) % Slope from 44%/min 95% 89% 100%

linear estimate (19/20) (8/9) (11/11)

Hanna et al. (1993) % Slope from 60%/min 89% 67% 100%

linear estimate (8/9) (2/3) (6/6)

Reddick et al. (1996) DVM: max slope 1.8/min 89% 86% 100%

and amplitude (17/19) (12/14) (5/5)

Kawai et al. (1997) % Slope from 60%/min 91% 86% 100%

Ongolo-Zogo et al. (1999) % Slope from Increase 92% 83% 100%

a Accuracy was calculated using retrospectively selected criteria in all cases, and therefore represents predicted, rather than measured, accuracy.

(10)

duces two different factor images from the DCE-MRI images: a “vascular factor” corresponding to early uptake of the contrast (ﬁ rst 1–2 min) which occurs in viable tumor, physes, and arteries; and an “accu- mulation factor” corresponding to delayed accu- mulation as the result of inﬂ ammation and edema.

Note that these factor images correspond well with the two factors (slope and maximum enhancement) used to compute the DVM measure. They also form the basis of a “DCE-MRI light” method suggested by Ma (1999), in which images are acquired at contrast injection, and at 30, 60, 90 and 120 s post-injection. In this method, the difference between the images at zero and 30 s post-contrast injection would correspond to the FAMIS “vascular factor” image and the difference between the images at zero and 120 s would corre- spond to the FAMIS “accumulation factor” image.

A group including Van der Woude, Verstraete and Bloem developed another qualitative method with similarities to the FAMIS approach. They identi- ﬁ ed three phases of the DCE-MRI intensity-time curves: ﬁ rst pass or wash-in of contrast, equilibra- tion, and wash-out (van der Woude et al. 1995, 1998 a,b; Verstraete et al. 1996; Verstraete and Lang 2000). A point-to-point correlation between DCE-MRI and histology revealed that enhancement within 3−6 s after the start of arterial enhancement corresponded to viable tumor and could discrimi- nate between reactive tissue and tumor. Note that this requires at least ultrafast or so-called turbo T1-weighted sequences if multiple slices are to be dynamically imaged (Shapeero and Vanel 2000).

Visual inspection of such fast “ﬁ rst-pass” images easily discriminated highly perfused viable regions

− those enhancing at <6 s after contrast arrived in the feeding artery − from edema, normal tissue and necrosis. Response was assessed by the disappear- ance of these bright regions on the ﬁ rst-pass images at the end of therapy, compared to those from the beginning.

In spite of a notable variety in methods of ana- lyzing clinical DCE-MRI data, studies of substantial series of patients with osteosarcoma for response to initial therapy have consistently reported that regions which enhanced brightly and/or reached near maxi- mum enhancement rapidly (within the ﬁ rst 60−90 s after contrast injection) were correlated with blood vessels and viable tumor regions. For more than a decade, DCE-MRI of osteosarcoma has been per- formed and validated against histological analyses of en bloc resections following initial chemotherapy.

The results of these studies, regardless of the analysis procedure or measure, have demonstrated accuracies

of approximately 90% for discrimination of response.

The parametric maps produced by pixel-by-pixel analysis methods have enabled the detection of resid- ual disease as small as 3–5 mm

²

(Egmont-Petersen et al. 2000; Hanna et al. 1993). Residual osteosarcoma tends to coalesce in nodules, a pattern which facili- tates its detection by DCE-MRI. However, in the rare instances when the osteosarcoma exists as widely scattered individual cells, no MRI method should be expected to provide a satisfactory assessment of the amount of viable tumor.

This process of rigorous validation of DCE-MRI in osteosarcoma has allowed radiologists to distinguish with increased conﬁ dence residual viable tissue from surrounding edema, normal tissue and necrosis.

However, there are two major pitfalls in interpreting results. With most DCE-MRI analysis methods, fast contrast uptake into newly vascularized tissue, such as immature granulation tissue, may appear as viable tumor. Thus reactive tissue marrow signal changes due to GCSF treatment, for example, can be mistaken for tumor. On the other hand, tumor tissue with low vascularity, such as chondroblastic areas, may appear necrotic.

13.2.1.4

Clinical Application of DCE-MRI to Predict Disease-Free Survival

In most studies, the percentage of tumor necrosis induced by preoperative chemotherapy has been predictive of disease-free survival (Picci et al. 1994;

Hudson et al. 1990; Meyers et al. 1992; Glasser et al. 1992; Provisor et al. 1997). There is gener- ally excellent agreement in the results from differ- ent researchers when DCE-MRI-based response measures are correlated with histologic measures for bone sarcoma. However, histologic measures of tumor necrosis obviously cannot be used at the time of diagnosis − the optimal point for identifying patients at increased risk of recurrence − and is appli- cable only to patients with resectable tumors after the completion of preoperative chemotherapy.

A recent retrospective study of DCE-MRI exami- nations at the time of presentation and at the com- pletion of preoperative chemotherapy was conducted for 31 patients who received protocol-based therapy for non-metastatic osteosarcoma of the extremities (Reddick et al. 2001). All DCE-MRI examinations were analyzed with a two-compartment pharmaco- kinetic model, and results were compared with dis- ease-free survival. Disease-free survival was deﬁ ned as the interval from complete surgical resection to

(11)

treatment failure or most recent follow-up. The 2-year estimated disease-free survival in this patient cohort was 76.1%+8.0% (SE). At the time of analysis, there had been seven relapses, which occurred 5 months to 4.5 years after surgery. Follow-up intervals ranged from 5 months to 6.0 years with a median of 2.1 years.

A total of 17 patients were responders by histologic grading of necrosis, while the remaining 14 patients failed to achieve the required 90% necrosis at resection.

For the 31 patients, the median measure of regional contrast access (k

ep

) was 1.167 min

^–1

at completion of preoperative chemotherapy (mean = 1.25 min

^–1

, standard deviation = 0.40 min

^–1

, range = 0.34 to 2.54 min

^–1

). Histologic assessment of response, based on the degree of necrosis in the resected lesion, was not found to be a prognostic factor of disease-free survival in these patients (p = 0.884), using a Cox proportional hazards model. However, the estimated coefﬁ cient for k

ep

at completion of preoperative che- motherapy was signiﬁ cantly correlated to treatment outcome, in a Cox proportional hazards model, with lower regional access (lower k

ep

) predicting improved outcome. Neither k

ep

at presentation and nor the change in k

ep

during therapy were signiﬁ cantly cor- related with outcome.

While k

ep

at completion of preoperative chemother- apy was signiﬁ cantly correlated to treatment outcome, it is preferable to assess patient risk at presentation.

Since lower regional access after preoperative therapy was predictive of improved outcome, the relationship between regional access at presentation and change in regional access during therapy was investigated.

The linear regression of change in regional access as a function of regional access at presentation had an R- square of 0.85 and is shown in Fig. 13.3

The inverse relationship between regional access (k

ep

) after initial therapy and improved disease-free survival estimates is consistent with the hypothesis that the transfer rate of low-MW MR contrast agents between the vasculature and the extracellular ﬂ uid acts as a surrogate measure of drug delivery. A small value for the k

ep

after preoperative chemotherapy may indicate that a large proportion of the tumor tissue is necrotic and its perfusing microvasculature is greatly reduced. The regional access measure k

ep

calculated by pharmacokinetic analysis of the DCE-MRI is a function of both blood ﬂ ow and vascular permeabil- ity of vessels. Tumors with a poorer response to che- motherapy would have larger viable regions, which are highly angiogenic and therefore more permeable to the contrast agent, and such regions would result in a larger value for regional access at the end of therapy. In contrast, larger values for regional access at the time of diagnosis would be expected to corre- spond with better delivery of the contrast agent, and by inference, improved delivery of low-MW drugs.

These larger k

ep

values were also positively related to response to chemotherapy and to longer disease-free survival estimates, although the latter did not reach signiﬁ cance in this study.

13.2.2 Ewing’s Family Tumors (EFTs)

The Ewing’s family of tumors includes Ewing’s sar- coma and peripheral primitive neuroectodermal tumor (PNET). Classical Ewing’s sarcoma and PNET are now identiﬁ ed as the same tumor with variable differentiation, and are treated similarly. Although much more rare than osteosarcoma, Ewing’s family tumors are the second most common malignant osseous tumors in children and adolescents. A total of 70% of these tumors occur in patients under the age of 20, with a male preference (Grier 1997). The natural history of Ewing’s family tumors appears to be the same in the adult and the pediatric popula- tion, although pediatric Ewing’s family tumors have been studied much more extensively (Fizazi et al.

1998; Baldini et al. 1999). An interesting feature of these malignancies is that they are extremely rare in patients of African or Asian ancestry, but have similar incidence in other populations (Grier 1997).

The collaborative studies organized by the world- wide pediatric oncology community have greatly improved the outlook for patients with Ewing’s family tumors. With modern therapy, more than 60%

of patients with local EFT can be cured. However, new

Fig. 13.3. Scatter plot and linear regression analysis demon- strating change in effective regional access (k_ep) during preoperative chemotherapy as a signiﬁ cant function of k_ep at the time of presentation. Larger effective regional access at the time of presentation corresponds to greater decrease in kep during therapy

(12)

therapies for these tumors are still urgently needed, particularly for patients who present with metas- tases or develop them on therapy. The outcome for the group with disseminated disease is still poor – a recent study ﬁ nds they have only a 22% relapse-free survival rate at 5 years post diagnosis (Cotterill et al. 2000).

Ewing’s family tumors belong to the group of small round cell tumors, which also includes neuro- blastoma, rhabdomyosarcoma, and non-Hodgkin’s lymphoma. Both Ewing’s sarcoma and PNET are characterized by the same histochemical staining profi le and demonstrate one of several reciprocal genetic translocations: t(11;22); t(21;22) and t(7;22) (Ambros et al. 1991). Each of these reciprocal trans- locations fuses part of the EWS gene on chromo- some 22 with a transcription factor belonging to the ETS family. This unique translocation is the focus of research to design targeted therapies for these malig- nancies. A reverse transcriptase-polymerase chain reaction (RT-PCR) assay based on the Ewing’s trans- location fi nds tumor cells in patients’ bone marrow samples at levels well below those detectable even by microscopic examination. There are ongoing trials to determine the predictive signifi cance of RT-PCR fi ndings at diagnosis (Grier 1997).

Ewing’s-PNET can develop in any bone in the body, with ﬂ at and long bones almost equally represented.

Approximately 50% of EFT occur in the extremi- ties. EFT occur predominantly in the diaphyses, whereas osteosarcoma is more commonly found in the metaphyses. They can occur as soft tissue tumors also, and in this form have no speciﬁ c distinguishing features on MRI. Negative prognostic factors include metastases at diagnosis, large tumor size/volume, and age over 17 years. The percentage of patients present- ing with metastases is usually said to be around 25%

(Green 1985) , although in two series of adult patients, 29% presented with metastases (Baldini et al. 1999;

Fizazi et al. 1998). In a large cooperative study, the incidence of metastases detected at presentation has increased signiﬁ cantly in the post-1991 period, com- pared to the 1980s, probably due to improved imag- ing in the latter period (Hense et al. 1999).

Positive prognostic factors include absence of metastases at diagnosis, location of tumor, young age, and good radiological and/or pathological response to pre-surgical chemotherapy (Cotterill et al. 2000;

Abudu et al. 1999). Centrally located tumors have worse prognosis. However, the correlation between tumor location and outcome is absent in some stud- ies, and it may not be an independent predictor. The importance of obtaining a good response to initial

chemotherapy was underlined by one recent study which tracked actual administered doses of chemo- therapeutic agents: those patients receiving higher actual doses were at ten-fold lower risk of metastatic recurrence than those patients who received less dose (Delepine 1997).

Current measures of response are based on radio- logical or pathological identiﬁ cation of percent necro- sis induced by initial chemotherapy in the tumor (Wunder et al. 1998; Aparicio et al. 1998; Abudu et al. 1999; Paulussen et al. 2001; Jenkin et al. 2002).

However, for EFT the criterion of response does not override the negative effect of metastases at presenta- tion: children with metastatic Ewing’s family tumors have poor chance of survival even when complete response to pre-surgical chemotherapy is observed (Cotterill et al. 2000). Therefore, in studies of response, it is important to analyze the local-disease and metastatic groups separately.

Because disseminated disease is such a key factor in staging these tumors, appropriate imaging proto- cols for identifying metastases are a critical part of staging for these tumors. EFT tend to metastasize to lung and bone marrow. Imaging protocols for staging therefore include radiographs and CT of the chest, plus bone scintigraphy and bone marrow biopsy, as well as clinical data, radiographs and MRI of the pri- mary tumor (van der Woude et al. 1998b; Ma 1999;

Grier 1997) .

13.2.2.1

Rationale for DCE-MRI of EFT

MRI is the preferred modality for therapy planning in Ewing’s family tumors, for many of the same reasons it is preferred for osteosarcoma therapy planning.

Over 50% of EFT occur in the extremities, and limb- salvage surgery is preferred over amputation when possible. MRI shows excellent accuracy in depicting the extension of osseous osteosarcoma and Ewing’s tumors to the physis and epiphysis (San Julian et al. 1999).

As with osteosarcoma, the gold standard for response of EFT to pre-operative chemotherapy is histological analysis for the presence of necrosis in resected tumor. The degree of response can be critical in the decision to use limb-sparing surgery.

The standard threshold separating good respond- ers from poor responders in EFT is >90% necrosis (<10% viable tumor) (Rosen et al. 1982; MacVicar et al. 1992). The extra-medullary component of EFT is predominantly cellular and, unlike osteosarcoma, may show a rapid decrease in volume in response to

(13)

chemotherapy. However, the use of volume change as a criterion for response is fraught with complications even in EFT. An increase in edema or the occurrence of hemorrhage in a responding tumor can be mis- taken for non-response. On the other hand, a tumor showing signiﬁ cant shrinkage may still contain active residual tumor cells that make up more than 10% of its volume. Radiography, CT, scintigraphy and static contrast-enhanced MRI have not proved able to determine tumor response accurately (Shapeero and Vanel 2000; Abudu et al. 1999).

13.2.2.2

Clinical Application of DCE-MRI to Assess EFT Response

As with osteosarcoma, the percentage of necrosis induced in Ewing’s family tumors by pre-surgical chemotherapy is strongly and positively associated with good patient outcome (both disease-free and overall survival) (Abudu et al. 1999). EFT are rarer than osteosarcoma, and much less attention has been devoted to DCE-MRI studies of these tumors. For over a decade, several European groups have pub- lished studies assessing the efﬁ cacy of DCE-MRI in measuring response to initial therapy, by comparing DCE-MRI measures of percent necrosis with point- by-point histologic measures (Verstraete et al.

1994 a–c; van der Woude et al. 1994, 1995, 1998 a,b;

Egmont-Petersen et al. 2000; Shapeero and Vanel 2000) . However, most reports have grouped the response of osteosarcoma and Ewing’s family tumors together, or have covered a mixed group of bone and soft-tissue sarcomas. As has been noted, this can cause confusion because the two tumors differ in their clinical course, radiologic appearance and, not least, in their histologic characteristics (MacVicar et al. 1992). Residual tumor in osteosarcoma tends to coalesce in nodules that are larger than the scat- tered, microscopic clusters of residual tumor found distributed throughout the soft and bony portions of EFT (MacVicar et al. 1992). This makes it more difﬁ - cult to determine the percentage of residual tumor in Ewing’s sarcoma, compared to osteosarcoma. These researchers have, however, succeeded in showing that DCE-MRI can detect residual tumor volumes of 3–5 mm

²

.

As EFT presents a more difﬁ cult imaging target for assessing response to initial chemotherapy, it is nec- essary to validate DCE-MRI for EFT separately from osteosarcoma. This has not yet been done. Only two studies have focused solely on DCE-MRI for measur- ing response in EFT.

The ﬁ rst is a small study of eight consecutive patients with Ewing’s sarcoma, all located in the legs (Egmont-Petersen et al. 2000). No information was given about metastatic disease at diagnosis. All eight underwent DCE-MRI at 0.5 T pre-operatively to assess local response to chemotherapy, and a path- ological specimen was available as a gold standard for the imaging results. DCE-MRI studies were per- formed at 0.5 T with a temporal resolution of 3.3 s per image plane and a spatial resolution of 0.61–3.0 mm

²

in plane by 8-mm section width. To avoid sacrifi c- ing spatial resolution in the analysis and to test sev- eral pharmacokinetic modeling approaches, the researchers fi tted each image pixel to pharmacoki- netic models, a technique used by others for osteosar- coma (see above), and produced parametric images for wash-in, wash-out, maximum enhancement, and arrival time of contrast. They carefully matched the histologic macroslice to the parametric images, so that they could make a region mask for viable tumor, which was matched to the parametric images in the same plane, and could calculate statistical measures of correctness for each model in each patient. The two models differed in that one assumed a single global arrival time for the contrast at all regions of the tumor, and the other calculated the arrival time for each pixel as a parameter of the fi t.

This study arrived at three salient results. First, comparisons with histology showed that their wash- in parametric image best identiﬁ ed the remnants of viable tumor in these patients. Their wash-in param- eter m

1

is proportional to the k

ep

of the standard ter- minology (Tofts et al. 1999). The overall sensitivity to tumor ranged from 0.33 to 0.77, with speciﬁ cities for necrosis of 0.58–0.99. Second, they demonstrated convincingly that the reliability of the wash-in param- eter

,

or k

ep

, is a strong function of the average size of the residual tumor. Simple pharmacokinetic model- ing will yield a k

ep

that is reliable only when residual

“islands” of tumor exceed a certain size – in this study, about 50−250 mm

³

. This threshold, however, is not an absolute number but a function of the SNR, the partial-volume effects, and the problems inherent in comparing microscopically thin histologic slices with 4-mm image planes.

Finally, they found that the more complex model (with arrival time as a parameter, not a constant) gave the most accurate fi t, but it only worked well when SNR was suffi ciently high. They found that the range of variation of the fi tted arrival times was +10 s within a typical region of interest – a very big range, given that the entire fi rst pass of contrast through the circulation is of the order of 10 s. Moreover, they

(14)

determined that by making the model more sensi- tive and therefore more accurate in computing the slope or wash-in parameter of the time–intensity curve, the model also became more sensitive to the noise in the data and therefore less robust. This is an excellent example of the sort of problem that can arise when ﬁ tting noisy data. Designing sophisticated algorithms to work around this apparent trade-off between accuracy and robustness is an active area of research in applying pharmacokinetic models to DCE-MR images.

The second, more recent study looked at 21 patients with a very different distribution of EFT: only 43%

were in extremities and a large fraction (43%) had metastatic disease at diagnosis (Miller et al. 2001).

The single-slice, T1-weighted spoiled gradient echo sequence used had a spatial resolution of 2.4–3.8 mm

²

in plane and 6–10 mm slice thickness. Temporal reso- lution was 7.6–14 s/image plane. They found that no DCE-MRI parameter showed a signiﬁ cant correla- tion with disease-free survival (DFS) and progres- sion-free survival (PFS), but that measures of tumor volume from routine MRI did correlate with survival.

However, it should be noted ﬁ rst, that this study has an unusually high fraction of patients with metastatic disease, and second, that the 12 patients with local disease were not analyzed separately from the nine with metastases at diagnosis. It is well established that response to standard initial chemotherapy does not correlate with outcome for patients with metastatic disease at or during therapy; these patients do very poorly regardless of response. The fact that nearly half the patients in this study had metastases would be expected to confound the correlation between any response measure and survival. The correlation these researchers found between tumor size at diagnosis and survival measures is probably linked to the pres- ence of metastases. Several studies have reported that the risk of metastases increases with tumor size, and in fact the risk of metastases correlated with tumor size in this study as well.

At week 8 (just prior to surgery), the DCE-MRI parameters ICAR (representing initial contrast uptake rate) and the lumped parameter DVM (which includes both the uptake rate and the maximum enhancement) were beginning to approach sig- niﬁ cance (p values of 0.05 and 0.06, respectively). It would be of interest to revisit the local-disease popu- lation in this study in order to determine the corre- lation between DCE-MRI and survival. This analysis could not be performed originally because statistical conditions for survival analysis were not met for this group.

13.2.3 Cartilaginous Tumors: Enchondroma vs. Chondrosarcoma

A study of DCE-MRI in 37 patients with cartilaginous tumors (eight enchondroma, 11 osteochondroma and 18 chondrosarcoma) (Geirnaerdt et al. 2000) looked at the utility of DCE-MRI in detecting low-grade chon- drosarcoma. Enchondroma and osteochondroma are benign lesions that do not require surgery. In con- trast, chondrosarcoma are malignant lesions whose only curative therapy is surgical resection. Although high-grade chondrosarcoma have distinctive malig- nant radiographic features, differentiation between benign and low-grade malignant cartilaginous tumors is a histological and radiological challenge. The clini- cal signiﬁ cance of this challenge is increased by the facts that low-grade chondrosarcoma are the more common, and that intralesional procedures increase the risk for local high-grade recurrence and the devel- opment of metastases, with a corresponding decrease in overall survival for these patients.

DCE-MRI parameters were correlated with the diagnosis from surgical or biopsy material. Although the data spanned two imaging systems (0.5 T and 1.5 T), with differences in the DCE-MRI timing and coverage, the investigators were able to show that a DCE-MRI pattern of very early exponential enhance- ment – e.g., tumor enhancement whose rise paralleled that of a nearby artery, but was delayed no more than 10 s after the arterial rise – correlated strongly with the diagnosis of chondrosarcoma. Early enhancement (deﬁ ned in this study as lesion enhancement starting

<10 s after arterial enhancement) was observed in all chondrosarcoma, none of the enchondroma, and in osteochondroma only in children with unfused growth plates. Using the early enhancement as a sign of malignancy, DCE-MRI identifi ed chondrosarcoma with a sensitivity of 89% and a specifi city of 84%. These detection rates are not very satisfactory, and it is not clear that DCE-MRI has any benefi ts over static con- trast-enhanced imaging for distinguishing between enchondroma and low-grade chondrosarcoma.

However, this example illustrates the importance of temporal resolution in certain DCE-MRI appli- cations. In an earlier study, investigators were not able to discriminate between enchondroma and low grade chondrosarcoma, most probably because of insufﬁ cient temporal resolution (20 s per image) (Erlemann et al. 1989). However, with a temporal resolution of 3 s per image, Verstraete et al. (1994a) were able to detect differences in the rate (slope) of early enhancement between malignant (two chon-

(15)

drosarcoma) and benign tumors (four osteochon- droma, seven enchondroma). The importance of detecting early enhancement at <10 s is conﬁ rmed in a recent study (Geirnaerdt et al. 2000).

13.3 Primary Soft Tissue Sarcomas

Soft tissue sarcomas are rare malignancies, although the reported incidence is increasing (Shapeero et al.

2002). They are also a very heterogeneous group of cancers, whose pathologic classiﬁ cation has evolved dramatically over the past 30 years. The lineage and genetic features of some malignancies in this group are well understood (e.g., rhabdomyosarcoma), while for others even the cell of origin is a matter of debate.

The classiﬁ cation of soft tissue sarcomas thus is something of a work in progress.

Historically, the most common tumor type within this group has been the malignant fi brous histiocy- toma (MFH). However, the validity of the MFH clas- sifi cation is being challenged. Two major functions of classifi cation are to associate cancers with specifi c cell lineages and to identify prognostic groups. An argu- ment has been made that many tumors identifi ed as subtypes of MFH are in fact soft tissue malignancies of distinctly different lines of differentiation, such as leiomyosarcoma or liposarcoma. In a retrospec- tive histologic assessment of MFH subtypes in 100 consecutive patients, 84 could be assigned to specifi c lines (Fletcher et al. 2001). The most common dif- ferentiated sarcomas found were myxofi brosarcoma (29), myogenic sarcoma (30), and myofi broblastic sarcoma (11). For oncologists, the importance of this discussion is the assertion that those tumors now classifi ed as pleomorphic MFH that show myogenic differentiation are signifi cantly more malignant, with shorter time to metastasis and worse outcome, than non-myogenic MFH of comparable stage (Fletcher et al. 2001). This has serious implications for the selection of study populations, and should be kept in mind in the design and interpretation of clinical studies of soft tissue sarcoma.

MRI is the most important imaging modality in the evaluation of soft tissue tumors, again because of its superior soft tissue contrast and the clarity with which it shows the tumor and its relation to tissue anatomic compartments. As is the case for bone sar- comas, biopsy is required for diagnosis of soft tissue malignancy (Crim et al. 1992), although appropriate imaging can narrow the differential.

MRI is critical in staging soft tissue neoplasms (Verstraete and Lang 2000; Berger et al. 2000;

Noria et al. 1996; Rao 1993), and should occur before any biopsy or resection is attempted. A case can be made that DCE-MRI is useful in targeting biopsies in some cases, e.g., identiﬁ cation of liposar- comatous degeneration within lipoma (Shapeero et al. 2002), and there is evidence that DCE-MRI can improve accuracy of diagnosis for leiomyosar- coma arising within uterine leiomyoma (see below).

Improvements in the accuracy of biopsy planning and resection are needed, because incomplete resec- tion, including contamination from intralesional procedures, is currently the most common cause of amputation for extremity soft tissue tumors (Noria et al. 1996; Kaste et al. 2002; Mankin et al. 1996). The problem of detecting residual tumor after incomplete resection is therefore a major issue with these lesions, and has not received the attention it deserves (Kaste et al. 2002; Noria et al. 1996). However, a number of studies show that static MRI is not sensitive to microscopic disease (Kaste et al. 2002; Mankin et al.

1996), nor is any MRI method, including DCE-MRI, able to discriminate between benign and malignant tissues with sufﬁ cient sensitivity (Crim et al. 1992;

Verstraete et al. 1996; May et al. 1997). Conse- quently, decisions regarding resection margins cannot be based on MRI alone.

13.3.1 Clinical Application of DCE-MRI to Distinguish Malignant from Benign Soft Tissue Tumors

There have been several very recent reports of the utility of DCE-MRI in specifi c diagnostic questions and DCE-MRI-guided biopsy. The fi rst two addressed challenges in distinguishing specifi c soft tissue malig- nancies from benign tumors.

The ﬁ rst study looked at the DCE-MRI character- istics of synovial sarcoma (van Rijswijk et al. 2001).

Synovial sarcoma classically occurs in soft tissues of extremities, especially near large joints, but can occur anywhere in the body, distant from joint spaces. It typ- ically affects adults in the fourth decade, but half the cases are in children and adolescents. Patients with localized synovial sarcoma have an excellent prog- nosis, and most are long-term survivors. Synovial sarcomas must be distinguished from hematoma or benign cystic masses, with which they are frequently confused (McCarville et al. 2002). Based on DCE- MRI studies of ten biopsy-proven synovial sarcoma, the most consistent DCE-MRI characteristic was